Y CombinatorMichael Kratsios: Inside the White House's AI Strategy
Y Combinator
From Silicon Valley to the White House 0:08
Michael Kratsios introduces himself as director of the White House Office of Science and Technology Policy and a special adviser to President Trump, having previously served as the country's chief technology officer in the first Trump term and later as chief operating officer at Scale AI. He explains that he moved between government and the private sector because he believes American leadership in emerging technologies is one of the most critical questions of the era, and that no workplace offers bigger problems to solve than the federal government, even compared to the largest tech companies.
An early fascination with technology 3:02
Kratsios traces his interest in technology back to obsessively following Steve Jobs keynotes and remembering the excitement of the first iPhone in college, even though he was not an engineer. He says the pieces came together during nearly seven years working for Peter Thiel in San Francisco between 2010 and 2017, where he saw repeatedly that government regulation shaped whether companies like Lyft or SpaceX could succeed, which pushed him toward government work focused on unlocking rules for drones, autonomous vehicles, and deliveries.
Why tech policy is so scattered 4:30
He points out that, unlike health or defense, there is no single federal technology department, so decisions on tech issues are spread across many agencies covering national security, commercial interests, and scientific research. Making good policy, he says, means the White House has to pull all these agencies together, using the example of drone rules needing both FAA approval and assurance from nuclear security officials that drones will not fly over sensitive sites.
Clarifying the stance on open source 6:01
Responding to a week of speculation that the administration might restrict open-weight AI models, Kratsios says this concern does not match reality. He notes that support for open source was stated on page one of the AI Action Plan released in July, a strategy he co-authored with David Sacks and Secretary Rubio, and that a vibrant ecosystem needs both closed and open models working together. He adds that public pushback from the startup community during moments like this is valuable because it lets policymakers hear directly from industry before decisions are made.
Balancing competing views on AI 8:30
Kratsios describes Washington's view of AI as an issue that touches data centers, healthcare, job loss, and productivity all at once, which makes it hard to separate these threads the way a single industry room might. He notes that Americans generally dislike having data centers nearby even though massive compute capacity is needed, so government has to balance public sentiment against technical necessity, drawing on input from builders to get details right.
Avoiding rigid rules in a fast-moving field 10:00
Asked how to regulate something that changes every six months, Kratsios warns against setting firm red lines, pointing to the European Union's AI Act, which was finalized before ChatGPT existed and now forces newer large language models to comply with rules written for a different era. He also cites the Biden administration's fixed compute threshold as an example of a rigid rule that does not age well, explaining that his team tries to keep policy flexible enough to move with the technology's frontier.
Weighing real risks against overblown fears 11:32
On AI risk, Kratsios names cybersecurity as a genuine near-term concern, describing how labs had to build safeguards into recent model releases to limit dangerous capabilities while preserving the same models' usefulness for defensive cybersecurity work. He argues that fears about biological risk from AI, loudly raised around 2021 and 2022, have proven somewhat overblown in practice, though he still sees value in building proper testing and evaluation infrastructure as models advance.
Making room for little tech 13:02
Addressing the dominance of large tech companies in policy conversations, Kratsios says the administration works to include smaller companies through mechanisms like public requests for comment, and he points to the push for federal preemption of a state-by-state patchwork of AI laws as a concrete example. He argues that big companies like Google can afford armies of lawyers to navigate fifty different state rules, but smaller companies cannot, so the president has called for one national standard to make it easier for anyone to build an AI company. He also credits continuity in antitrust cases against major tech firms as part of protecting space for smaller players.
Cutting regulatory barriers to innovation 18:31
Kratsios distinguishes between technologies that are born free, meaning no regulations yet exist and should be preserved from new rules, like the early internet, and those born in captivity, meaning they require government approval before they can be commercialized, like commercial drone flight. He cites Boom Supersonic as an example of a born-in-captivity technology held back by an outdated speed-based noise rule, and says his office focuses on removing exactly these kinds of regulatory barriers so promising technologies can reach the market.
How the White House policy machine works 21:00
Kratsios explains that federal policy is organized into four councils inside the White House covering national security, domestic policy, economic policy, and science and technology, each led by a senior adviser who coordinates the relevant agencies. His own portfolio includes AI, quantum computing, civil nuclear energy, biotech, and space, each staffed by small teams, such as three people handling space policy who bring together NASA and the Department of War. He describes a typical day as a mix of stakeholder meetings, hands-on policy work like pushing the FAA to change drone rules, and public engagement to explain the administration's priorities and gather feedback.
How Policy Disputes Reach the President 25:00
When agencies disagree on a tech or AI policy question, the disagreement moves up the chain rather than stalling. Kratsios describes a bottom-up process: staff try to resolve disputes first, and if they cannot, the issue rises to more senior officials, and if it still cannot be resolved, it eventually reaches the president himself. The goal is to keep the number of decisions reaching the president small, since his time is limited and should go toward the most consequential national priorities. Only the highest-stakes AI or technology questions are brought directly to him for a final call.
A New Science Report Modeled on 1945 26:32
Kratsios discusses a roughly 150-page report, titled something like "science, a new golden age," released that week. He traces its roots to 1945, when FDR asked his science adviser Vannevar Bush how to organize the science ecosystem after World War II, producing the famous report "Science, the Endless Frontier," which argued government should fund early-stage basic research the private sector wouldn't. Back then the federal government funded nearly 70 percent of R&D; today the private sector and philanthropy fund about 70 percent, with roughly equal footing in basic research at universities. President Trump asked Kratsios a similar question, and the resulting report argues AI is transforming scientific discovery in fields like materials science, pharmaceuticals, and chemistry, and that the roughly 200 billion dollars the government spends yearly on R&D must adapt. The report envisions autonomous cloud labs running experiments, testing hypotheses, and reaching conclusions with minimal human involvement, alongside concrete measures like fast grants decided within a month and agency action plans due in 90 days. Kratsios also flags quantum information science as an underappreciated priority, noting Trump's new executive order and a goal for the Department of Energy to build a scientifically relevant quantum computer by 2028.
Congress, Intellectual Property, and Advice for Young People 34:01
Kratsios contrasts executive orders, which can be revoked when administrations change, with laws, which stick, and says the White House wants Congress to pass preemption legislation to avoid a patchwork of state AI rules. He also wants clearer statutory protection around AI and intellectual property, stating flatly that a model outputting Mickey Mouse is "not cool" since Disney should control that. He favors letting the market develop revenue-sharing deals with creators, citing Yelp's recent licensing deal with ChatGPT, rather than rushing to legislate. Closing, he encourages young technologists to consider government service, calling it slow and bureaucratic at times but uniquely fulfilling when results land, and invites attendees to reach out to the White House science and technology office.
AI-generated summary. It can be wrong or incomplete - check anything that matters against the original.
